OkkhorDiffusion / README.md
ahmedfaiyaz's picture
Update README.md
1963c01 verified
|
raw
history blame
2.24 kB
metadata
license: cc-by-nc-2.0
language:
  - bn
library_name: diffusers
pipeline_tag: text-to-image
inference: false

Okkhor Diffusion is a category of Denoising Diffusion Probabilistic Models designed to generate images of Bangla handwritten characters.This model cards corresponds to Okkhor-Diffusion trained on Banglalekha-Isolated dataset.Other variants will be uploaded soon.

Usage

from diffusers import DiffusionPipeline
from typing import List, Optional, Tuple, Union
import torch
pipeline = DiffusionPipeline.from_pretrained("ahmedfaiyaz/OkkhorDiffusion",custom_pipeline="gr33nr1ng3r/OkkhorDiffusion",embedding=torch.float16)
pipeline.to("cuda")
pipeline.embedding=torch.tensor([10-1]) # 'ও': 9
pipeline(batch_size=1,num_inference_steps=1000).images[0]

Character mappings

character_mappings = {
    'অ': 1,
    'আ': 2,
    'ই': 3,
    'ঈ': 4,
    'উ': 5,
    'ঊ': 6,
    'ঋ': 7,
    'এ': 8,
    'ঐ': 9,
    'ও': 10,
    'ঔ': 11,
    'ক': 12,
    'খ': 13,
    'গ': 14,
    'ঘ': 15,
    'ঙ': 16,
    'চ': 17,
    'ছ': 18,
    'জ': 19,
    'ঝ': 20,
    'ঞ': 21,
    'ট': 22,
    'ঠ': 23,
    'ড': 24,
    'ঢ': 25,
    'ণ': 26,
    'ত': 27,
    'থ': 28,
    'দ': 29,
    'ধ': 30,
    'ন': 31,
    'প': 32,
    'ফ': 33,
    'ব': 34,
    'ভ': 35,
    'ম': 36,
    'য': 37,
    'র': 38,
    'ল': 39,
    'শ': 40,
    'ষ': 41,
    'স': 42,
    'হ': 43,
    'ড়': 44,
    'ঢ়': 45,
    'য়': 46,
    'ৎ': 47,
    'ং': 48,
    'ঃ': 49,
    'ঁ': 50,
    '০': 51,
    '১': 52,
    '২': 53,
    '৩': 54,
    '৪': 55,
    '৫': 56,
    '৬': 57,
    '৭': 58,
    '৮': 59,
    '৯': 60,
    'ক্ষ(ksa)': 61,
    'ব্দ(bda)': 62,
    'ঙ্গ': 63,
    'স্ক': 64,
    'স্ফ': 65,
    'স্থ': 66,
    'চ্ছ': 67,
    'ক্ত': 68,
    'স্ন': 69,
    'ষ্ণ': 70,
    'ম্প': 71,
    'হ্ম': 72,
    'প্ত': 73,
    'ম্ব': 74,
    'ন্ড': 75,
    'দ্ভ': 76,
    'ত্থ': 77,
    'ষ্ঠ': 78,
    'ল্প': 79,
    'ষ্প': 80,
    'ন্দ': 81,
    'ন্ধ': 82,
    'ম্ম': 83,
    'ন্ঠ': 84,
}